WeSCoS Colloquium #05 “Improvement of resolution of meteorological data by deep learning considering physical laws and symmetry” By Yuki Yasuda (Specially Appointed Associate Professor, Global Scientific Information and Computing Center, Tokyo Institute of Technology)

WeSCoS Colloquium #05 “Improvement of resolution of meteorological data by deep learning considering physical laws and symmetry”
Reporter: Yuki Yasuda (Specially Appointed Associate Professor, Global Scientific Information and Computing Center, Tokyo Institute of Technology)
Date and time: April 19, 2023 (Wednesday) 13:00-　
overview:
In recent years, deep neural networks have been applied to fluid systems, and their usefulness has been confirmed in various problems. Many of them apply techniques based on image processing, but it is not obvious whether these techniques are valid for fluid systems. In this presentation, we discuss the physical validity of inference for neural networks that perform high-resolution meteorological data. The first half describes how to incorporate governing laws into learning to improve validity. In the second half, we discuss the conditions under which a neural network developed for image (color, scalar) processing can treat the velocity vector as a quantity with a “direction” from the point of view of equivariance.
Method: Zoom
*Pre-registration is not required.
Registration Form
·Language: Japanese
* WeSCoS Colloquium is a moonshot R&D project goal 8
A colloquium on the core research “Control Theory of Meteorological-Social Coupled Systems Supporting Social Decision-Making” will be open to the public.
*WeSCoS is an abbreviation for Weather-Society Coupling/Control System.